JP2019513246A - Training method of random forest model, electronic device and storage medium - Google Patents

Abstract

The present invention relates to a random forest model training method, an electronic device, and a storage medium. The training method of the random forest model is to analyze whether the model training control system satisfies the condition for performing the model training and to the random forest model when the condition for performing the model training is satisfied. If it is necessary to determine whether or not reconstruction training needs to be performed, and if it is necessary to perform reconstruction training on the random forest model, reconstruction on the random forest model using sample data is performed. Training, and performing corrective training on the random forest model using sample data, if it is not necessary to perform reconstruction training on the random forest model. The present invention can reduce the number of training of the random forest model, reduce the load on the system, and increase the efficiency of the system.

Description

"Cross-reference to related applications"
The present application is based on the Paris Convention and applied to the State Intellectual Property Office of China on March 13, 2017, under the Chinese Patent Application No. 2017101476983 entitled "Chinese Forest Patent Training Method and Model Training Control System". The priority of the application is claimed, and the entire application is incorporated herein by reference.

  The present invention relates to the technical field of machine learning, and more particularly, to a random forest model training method, an electronic apparatus, and a storage medium.

  A random forest in machine learning is a classifier including multiple decision trees, which trains and predicts sample data using a large number of decision trees, and a decision tree is a process of classifying data according to a series of rules. Point to. Currently, many companies that offer online services (eg, remote insurance coverage, remote damages claims, online asset management, etc.) have already introduced random forests in their service systems, and identified users by classification labels. , Based on the recognition results, targeted service presentation and processing.

  However, when sample data is newly provided for repetitive training to improve model recognition accuracy, the conventional technical solution simultaneously uses the existing sample data and the new sample data to generate the random forest model. We are trying to carry out the reconstruction training again. Reconstruction training refers to training for changing the structure of a decision tree in a random forest model. Such training methods generally perform reconstruction training when new sample data are provided, so the number of trainings is large, and data changes are particularly frequent in online services. In the situation where it is carried out, if the training is carried out frequently, the load on the system is increased, and as a result, the efficiency of the online service system and the progress of the online service are affected.

  An object of the present invention is to provide a training method, an electronic device and a storage medium of a random forest model in order to reduce the number of times of training of the random forest model, reduce the load on the system and improve the efficiency of the system.

A first aspect of the present invention provides a training method of random forest model,
The method is
Analyzing whether a condition for performing model training is satisfied by the model training control system;
Step S2 of determining whether the random forest model needs to be subjected to the reconstruction training if the condition for performing the model training is satisfied;
Performing reconstruction training on the random forest model using sample data when reconstruction training needs to be performed on the random forest model;
Performing corrective training on the random forest model using sample data, if it is not necessary to perform reconstruction training on the random forest model;
including.

A second aspect of the invention provides an electronic device,
The electronic device is
Processing equipment, storage equipment and model training control system,
The model training control system is stored in the storage device and includes at least one computer readable command;
The processing device executing the at least one computer readable command,
Analyzing whether a condition for performing model training is satisfied by the model training control system;
Step S2 of determining whether the random forest model needs to be subjected to the reconstruction training if the condition for performing the model training is satisfied;
Performing reconstruction training on the random forest model using sample data when reconstruction training needs to be performed on the random forest model;
Performing corrective training on the random forest model using sample data, if it is not necessary to perform reconstruction training on the random forest model;
Can be realized.

A third aspect of the present invention provides a computer readable storage medium,
At least one computer readable command is stored on the storage medium;
A processing device executing the at least one computer readable command;
Analyzing whether a condition for performing model training is satisfied by the model training control system;
Step S2 of determining whether the random forest model needs to be subjected to the reconstruction training if the condition for performing the model training is satisfied;
Performing reconstruction training on the random forest model using sample data when reconstruction training needs to be performed on the random forest model;
Performing corrective training on the random forest model using sample data, if it is not necessary to perform reconstruction training on the random forest model;
Can be realized.

  The present invention can set or limit conditions for performing random forest model training when classifying users using online service by using a random forest model, which affects the progress of online service. In addition to reducing the number of model trainings as well as not giving them a model training type, ie, if the conditions for performing model training are met, do you do the reconstruction training for the random forest model? By deciding whether to perform training or corrective training, it is possible to selectively train the random forest model, which greatly reduces the load on the system, increases the efficiency of the online service system, and improves the online service. Can proceed efficiently.

FIG. 1 illustrates an environment for implementing a preferred embodiment of the random forest model training method according to the present invention. 5 is a flow chart of a preferred embodiment of a method of training a random forest model according to the present invention. It is a detailed flowchart of step S4 in FIG. 1 is a schematic structural diagram of a preferred embodiment of a model training control system according to the present invention. FIG. 5 is a schematic structural view of a second training module in FIG. 4;

  The principles and features of the present invention will be described based on the examples and drawings as follows, but the listed examples are only for explaining the present invention and not for limiting the present invention. It is self-explanatory.

  FIG. 1 is an environment for implementing a preferred embodiment of a training method of random forest model according to the present invention, according to FIG. 1, said environment comprises an electronic device 1 and a terminal equipment 2. Among them, the electronic device 1 can exchange data with the terminal device 2 by an appropriate technology such as a network or short distance communication technology.

  The terminal device 2 is any electronic product that can perform man-machine interaction with the user by means of a keyboard, a mouse, a remote controller, a touch panel, or a voice control device, for example, a PC, a tablet PC, a smartphone, a personal digital assistant (Personal Digital Assistant, PDA), game console, Internet Protocol Television (IPTV), wearable terminal, etc., but not limited thereto.

  The electronic device 1 is a device capable of automatically performing numerical calculation and / or information processing in accordance with, for example, a preset or stored command. Examples of the electronic device 1 include a computer, a single network server, a server group including a plurality of network servers, or a cloud-type device including a large number of hosts and network servers based on cloud computing, and cloud computing , A kind of distributed computing, which refers to a large virtual computer consisting of a plurality of loosely coupled computer clusters.

  In the present embodiment, the electronic device 1 includes, but is not limited to, the storage device 11, the processing device 12, and the network interface 13 connected by the system bus so as to communicate with each other. FIG. 1 shows an electronic device 1 comprising a part 11, a part 12 and a part 13, but not necessarily all the parts shown, more or less as an alternative It may be implemented in

  Of these, the storage device 11 includes a memory and at least one type of readable storage medium. The memory provides a cache when the electronic device 1 operates, and the readable storage medium is, for example, a non-volatile storage medium such as a flash memory, a hard disk, a multimedia card, and a card type memory. In some embodiments, the readable storage medium is the internal storage means of the electronic device 1, for example the hard disk of the electronic device 1, and in another embodiment the non-volatile storage medium is external storage of the electronic device 1. For example, an external hard disk mounted on the electronic device 1, a smart media card (Smart Media (registered trademark) Card, SMC), a Secure Digital (SD) card, a flash card (Flash Card), or the like. In the present embodiment, an operating system and various application software in which a readable storage medium of the storage device 11 is generally installed in the electronic device 1, for example, a program code of the model training control system 10 in a preferred embodiment of the present invention. It is used to memorize etc. The storage device 11 is also used to temporarily store various data that has already been output or is output.

  In some embodiments, processing device 12 may include one or more microprocessors, microcontrollers, digital processors, and the like. The processing device 12 is generally used to control the execution of the operation of the electronic device 1, eg, control and processing associated with data exchange or communication with the terminal device 2. In the present embodiment, the processing device 12 executes program code stored in the storage device 11 or processes data, and is used, for example, to execute the model training control system 10 or the like.

  The network interface 13 may include a wireless network interface or a wired network interface, and the network interface 13 is generally used to establish a communication connection between the electronic device 1 and other electronic devices. In the present embodiment, the network interface 13 mainly connects the electronic device 1 and one or more terminal devices 2 to connect a data transmission channel between the electronic device 1 and one or more terminal devices 2. Used to establish a communication connection.

  The model training control system 10 includes at least one computer readable command stored in the storage device 11, and the processing device 12 executes the at least one computer readable command to implement each embodiment of the present invention. The training method of random forest model by example can be realized. As will be understood from the following description, the at least one computer readable command may be divided into different logic modules according to the difference in the function implemented by each part.

  In one embodiment, when the processing device 12 executes the model training control system 10, the model training control system analyzes the operation for satisfying the condition for performing the model training and the condition for performing the model training. If it is necessary to determine whether or not it is necessary to carry out the reconstruction training on the random forest model, and if it is necessary to carry out the reconstruction training on the random forest model, sample data Using the sample data to perform reconstruction training on the random forest model, and using the sample data to correct the random forest model if reconstruction training does not need to be performed on the random forest model And the action of performing

  FIG. 2 shows a flowchart of a preferred embodiment of the training method of random forest model according to the present invention, and according to FIG. 2, the training method of random forest model of this embodiment is limited to the steps in the flowchart of FIG. Alternatively, some of the steps in the flowchart of FIG. 2 may be omitted, or the order of the steps may be changed. The training method of the random forest model includes steps S1, S2, S3, and S4.

  Step S1: A model training control system analyzes whether the conditions for performing model training are satisfied.

  Model training includes reconstructive training and corrective training. The conditions for performing model training can be manually set flexibly in the model training control system, or default conditions preset in the model training control system can be applied.

  In the case of on-line services, the conditions for performing model training are based on the user's service data (e.g. performing a model training when the user's service data reaches a predetermined quantity) or based on the actual needs (e.g. , The model training control system operator performs model training by transmitting a model training execution command to the model training control system according to the actual needs, or clocking with a timer, and one model training is completed. Later, training is performed on the random forest model each time a predetermined time elapses.

Preferably, when analyzing whether the conditions for performing model training have been met,
The first quantity of service data of the user newly added in the period from the end of the last model training in the service system to the present time (for example, let the first quantity be the quantity of service data of 200 users) If the first quantity exceeds a first predetermined threshold, it is determined that the condition for performing model training is satisfied (reconstructional training or corrective training can be performed); And the condition for performing model training is not satisfied (reconstructive training or corrective training is not performed) if the number of
Alternatively, whether or not the model training command has been received is detected in real time or on a regular basis (for example, every 10 minutes), for example, the operator of the model training control system logs in to the system and operates on the model training operation screen. Incoming conditions for transmitting model training commands by clicking or triggering the “Model training” button on the model training operation screen, and for performing model training when the model training control system receives model training commands If the model training command is not received, the condition for performing model training is not satisfied (reconstruction training or correction training). Do not do training).

  Step S2: If the conditions for performing model training are satisfied, it is determined whether or not it is necessary to perform a reconstruction training on the random forest model.

  If the conditions for performing model training are satisfied, it is determined whether to perform reconstruction training or corrective training on a random forest model. Among these, determining whether or not the random forest model needs to be subjected to the reconstruction training is based on the service data of the user between the two reconstruction trainings (for example, twice). Based on the actual needs (for example, the operator of the model training control system responds to the actual needs), if the service data of the user during the reconstruction training exceeds the predetermined quantity, the reconstruction training is performed) Then, the model training control system transmits reconfiguration training execution command to perform reconfiguration training) and the like.

Preferably, when deciding whether or not the random forest model needs to be subjected to reconstruction training:
A second quantity of service data of the user newly added during the period from the end of the last reconstruction training in the service system to the current time (eg, the second quantity is the quantity of service data of 500 users) And if the second quantity exceeds a second predetermined threshold, it is considered necessary to perform reconstruction training on the random forest model, and the second quantity is a first predetermined threshold. Exceed the second threshold and perform corrective training on the random forest model,
Alternatively, it sends a request to a predetermined terminal (for example, a suitable electronic terminal such as a mobile phone, a tablet PC, a computer, etc.) as to whether or not it is necessary to perform reconfiguration training on the random forest model. After entering the training operation screen and transmitting a "model training" command, the model training control system transmits information inquiring whether or not to perform reconfiguration training on a predetermined terminal, and the terminal responds to the inquiry request from the terminal. When it is determined that the confirmation command that has been fed back is received, it is necessary to perform reconfiguration training on the random forest model, and a negative command that has been fed back from the terminal in response to the inquiry request is received or For example, when receiving no feedback from the terminal within 3 minutes) , A correction training with respect to the random forest model.

  Step S3: If it is necessary to carry out reconstruction training on the random forest model, conduct reconstruction training on the random forest model using sample data.

  Step S4: If the random forest model does not need to be reconstructed, sample training data is used to perform corrective training on the random forest model.

  In the present embodiment, the sample data includes existing sample data and newly added sample data. Reconstruction training includes deterministic training on variables of the random forest model and deterministic training on coefficients of the variables, and corrective training includes only deterministic training on coefficients of the variables of the random forest model. Among them, variables of the random forest model include, for example, the type of algorithm, the number of decision trees, the maximum depth of decision trees, and various data of leaves and nodes of decision trees. There are more system resources used for reconstructive training than system resources used for corrective training.

  Compared with the prior art, this embodiment can set or limit the conditions for training the random forest model when classifying the users who use the online service using the random forest model, Not only does it not affect the progress of online services, but it reduces the number of model trainings and selects the type of model training, that is, if the conditions for performing model training are met, the random forest model is re-created. By deciding whether to perform constitutive training or corrective training, it is possible to selectively train the random forest model, which greatly reduces the load on the system and allows the online service system to You can improve efficiency and promote online services efficiently.

  In one preferred embodiment, as shown in FIG. 3, based on the embodiment of FIG. 2, step S4 further includes steps S41 and S42.

  Step S41: The range of the coefficient value of the variable corresponding to each variable is specified based on the mapping relationship between the variable of the predetermined random forest model and the value range of the coefficient of the variable.

  Step S42: The value of the coefficient of the variable is acquired from the range of the value of the coefficient of the corresponding variable for each of the variables, and the random forest model is subjected to corrective training based on the coefficient of the variable whose value is acquired.

  In this embodiment, the range of values of variables and variable coefficients of the random forest model may be associated in advance to establish a mapping relationship, and the mapping relationship established by the association may be stored (for example, stored in the form of a table) ). Before training on the random forest model, determine the variables of the random forest model, and obtain the mapping relationship established by the stored association, and obtain the value range of the coefficient of the corresponding variable The accuracy of model training is ensured by acquiring the values of the coefficients of the variables limited to the obtained range of values, and the values of the coefficients of each variable of the random forest model are in the entire numerical range By avoiding training from training, model training can be effectively speeded up.

  FIG. 4 shows functional modules of a preferred embodiment of a model training control system 10 according to the present invention, and according to FIG. 4, in this embodiment, the model training control system 10 is divided into one or more modules. The one or more modules are stored in a memory and implemented by one or more processing devices to implement the invention. For example, according to FIG. 4, the model training control system 10 is divided into a detection module 21, a recognition module 22, a replication module 23, an installation module 24, and an activation module 25. The module in the present invention indicates a series of computer program command blocks for realizing a specific function, and is more suitable for describing the execution process of the model training control system 10 in the electronic device than the whole program.

  Among these, the analysis module 101 analyzes whether conditions for performing model training are satisfied.

  Model training includes reconstructive training and corrective training. The conditions for performing model training can be set manually in the model training control system 10 flexibly, or default conditions preset in the model training control system can be applied.

  In the case of on-line services, the conditions for performing model training are based on the user's service data (e.g. performing a model training when the user's service data reaches a predetermined quantity) or based on the actual needs (e.g. , The model training control system operator performs model training by transmitting a model training execution command to the model training control system according to the actual needs, or clocking with a timer, and one model training is completed. Later, training is performed on the random forest model each time a predetermined time elapses.

Preferably, specifically, the analysis module 10 is configured to calculate a first quantity (eg, a first quantity) of service data of a user newly added in a period from the time when the previous model training in the service system is finished to the present time. And the number of service data of 200 users), and if the first quantity exceeds a first predetermined threshold, the condition for performing model training is satisfied (reconstruction training or Corrective training can be performed), if the first quantity is less than or equal to a first predetermined threshold, the condition for performing model training is not satisfied (reconstructional training or corrective training is not performed),
Alternatively, whether or not the model training command has been received is detected in real time or on a regular basis (for example, every 10 minutes), for example, the operator of the model training control system logs in to the system and operates on the model training operation screen. Incoming conditions for transmitting model training commands by clicking or triggering the “Model training” button on the model training operation screen, and for performing model training when the model training control system receives model training commands If the model training command is not received, the condition for performing model training is not satisfied (reconstruction training or correction training). Do not do training).

  The determination module 102 determines, if the conditions for performing model training are satisfied, whether or not reconstruction training needs to be performed on the random forest model.

  If the conditions for performing model training are satisfied, it is determined whether to perform reconstruction training or corrective training on a random forest model. Among these, determining whether or not the random forest model needs to be subjected to the reconstruction training is based on the service data of the user between the two reconstruction trainings (for example, twice). Based on the actual needs (for example, the operator of the model training control system responds to the actual needs), if the service data of the user during the reconstruction training exceeds the predetermined quantity, the reconstruction training is performed) Then, the model training control system transmits reconfiguration training execution command to perform reconfiguration training) and the like.

Preferably, specifically, the determination module 102 determines the second quantity of service data of the user newly added during the period from the time when the previous reconstruction training in the service system is finished to the present time (for example, the second quantity Needs to perform the training on the random forest model if the number of services of 500 users is taken as the number of service data of 500 users and the second number exceeds a second predetermined threshold. And if the second quantity is greater than a first predetermined threshold and less than a second threshold, corrective training is performed on the random forest model,
Alternatively, it sends a request to a predetermined terminal (for example, a suitable electronic terminal such as a mobile phone, a tablet PC, a computer, etc.) as to whether or not it is necessary to perform reconfiguration training on the random forest model. After entering the training operation screen and transmitting a "model training" command, the model training control system transmits information inquiring whether or not to perform reconfiguration training on a predetermined terminal, and the terminal responds to the inquiry request from the terminal. When it is determined that the confirmation command that has been fed back is received, it is necessary to perform reconfiguration training on the random forest model, and a negative command that has been fed back from the terminal in response to the inquiry request is received or For example, when receiving no feedback from the terminal within 3 minutes) , A correction training with respect to the random forest model.

  The first training module 103 performs reconstruction training on the random forest model using sample data when reconstruction training needs to be performed on the random forest model.

  The second training module 104 performs corrective training on the random forest model using sample data, when it is not necessary to perform reconstruction training on the random forest model.

  In the present embodiment, the sample data includes existing sample data and newly added sample data. Reconstruction training includes deterministic training on variables of the random forest model and deterministic training on coefficients of the variables, and corrective training includes only deterministic training on coefficients of the variables of the random forest model. Among them, variables of the random forest model include, for example, the type of algorithm, the number of decision trees, the maximum depth of decision trees, and various data of leaves and nodes of decision trees. There are more system resources used for reconstructive training than system resources used for corrective training.

  Compared with the prior art, this embodiment can set or limit the conditions for training the random forest model when classifying the users who use the online service using the random forest model, Not only does it not affect the progress of online services, but it reduces the number of model trainings and selects the type of model training, that is, if the conditions for performing model training are met, the random forest model is re-created. By deciding whether to perform constitutive training or corrective training, it is possible to selectively train the random forest model, which greatly reduces the load on the system and allows the online service system to You can improve efficiency and promote online services efficiently.

  In one preferred embodiment, as shown in FIG. 5, based on the embodiment of FIG. 4, the second training module 104 includes identification means 1041 and training means 1042.

  The identifying unit 1041 identifies the range of coefficient values of the variable corresponding to each of the variables based on the mapping relationship between the predetermined random forest model variable and the range of the coefficient value of the variable.

  The training unit 1042 acquires the value of the coefficient of the variable from the range of the value of the coefficient of the corresponding variable for each of the variables, and performs correction training on the random forest model based on the coefficient of the variable whose value is acquired. Do.

  In this embodiment, the range of values of variables and variable coefficients of the random forest model may be associated in advance to establish a mapping relationship, and the mapping relationship established by the association may be stored (for example, stored in the form of a table) ). Before training on the random forest model, determine the variables of the random forest model, and obtain the mapping relationship established by the stored association, and obtain the value range of the coefficient of the corresponding variable The accuracy of model training is ensured by acquiring the values of the coefficients of the variables limited to the obtained range of values, and the values of the coefficients of each variable of the random forest model are in the entire numerical range By avoiding training from training, model training can be effectively speeded up.

  The preferred embodiments of the present invention have only been described above, and the present invention is not limited thereto, and any modifications, equivalent replacements, improvements, etc. which can be made within the spirit and scope of the present invention can be made. Are also included in the scope of protection of the present invention.

Claims (20)

The training method of the random forest model,
Analyzing whether a condition for performing model training is satisfied by the model training control system;
Step S2 of determining whether the random forest model needs to be subjected to the reconstruction training if the condition for performing the model training is satisfied;
Performing reconstruction training on the random forest model using sample data when reconstruction training needs to be performed on the random forest model;
Performing corrective training on the random forest model using sample data, if it is not necessary to perform reconstruction training on the random forest model;
The training method of the random forest model characterized by including. As said step S1,
Obtaining a first quantity of service data of a user newly added in a period from the end of previous model training in the service system to the current time, and the first quantity exceeds a first predetermined threshold, It is assumed that the conditions for performing model training are satisfied,
Alternatively, whether or not the model training command has been received is detected in real time or on a regular basis, and when the model training command is received, the condition for performing the model training is satisfied. Training method of random forest model. As said step S2,
A second quantity of service data of a user newly added in a period from the end of the last reconstruction training in the service system to the present time is obtained, and the second quantity exceeds a second predetermined threshold If it is necessary to perform reconstruction training on the random forest model,
Alternatively, a request for inquiring whether or not it is necessary to perform a reconfiguration training on the random forest model is transmitted to a predetermined terminal, and a confirmation command fed back from the terminal in response to the inquiry request is received. The training method of the random forest model according to claim 1, wherein it is necessary to perform reconstruction training on the random forest model. The reconstruction training includes deterministic training on variables of the random forest model and deterministic training on coefficients of the variables;
The method for training a random forest model according to claim 1, wherein the corrective training includes deterministic training on coefficients of variables of the random forest model. As said step S1,
Obtaining a first quantity of service data of a user newly added in a period from the end of previous model training in the service system to the current time, and the first quantity exceeds a first predetermined threshold, It is assumed that the conditions for performing model training are satisfied,
Alternatively, whether or not the model training command has been received is detected in real time or on a regular basis, and when the model training command is received, the condition for performing the model training is satisfied. Training method of random forest model. As said step S2,
A second quantity of service data of a user newly added in a period from the end of the last reconstruction training in the service system to the present time is obtained, and the second quantity exceeds a second predetermined threshold If it is necessary to perform reconstruction training on the random forest model,
Alternatively, a request for inquiring whether or not it is necessary to perform a reconfiguration training on the random forest model is transmitted to a predetermined terminal, and a confirmation command fed back from the terminal in response to the inquiry request is received. The training method of the random forest model according to claim 4, wherein it is necessary to perform reconstruction training on the random forest model. In the step S4,
Identifying a range of coefficient values of variables corresponding to each of the variables based on a mapping relationship between a predetermined random forest model variable and the range of coefficient values of the variables;
Acquiring the value of the coefficient of the variable from the range of the value of the coefficient of the corresponding variable for each of the variables, and performing correction training on the random forest model based on the coefficient of the variable whose value has been acquired;
The training method of the random forest model according to claim 4 characterized by including. An electronic device,
Processing equipment, storage equipment and model training control system,
The model training control system is stored in the storage device and includes at least one computer readable command;
The processing device executing the at least one computer readable command,
Analyzing whether a condition for performing model training is satisfied by the model training control system;
Step S2 of determining whether the random forest model needs to be subjected to the reconstruction training if the condition for performing the model training is satisfied;
Performing reconstruction training on the random forest model using sample data when reconstruction training needs to be performed on the random forest model;
If it is not necessary to carry out the reconstruction training on the random forest model, it is possible to realize the step S4 of performing the correction training on the random forest model using sample data. As said step S1,
Obtaining a first quantity of service data of a user newly added in a period from the end of previous model training in the service system to the current time, and the first quantity exceeds a first predetermined threshold, It is assumed that the conditions for performing model training are satisfied,
Alternatively, whether or not the model training command has been received is detected in real time or on a regular basis, and when the model training command is received, the condition for performing the model training is satisfied. Electronic devices. As said step S2,
A second quantity of service data of a user newly added in a period from the end of the last reconstruction training in the service system to the present time is obtained, and the second quantity exceeds a second predetermined threshold If it is necessary to perform reconstruction training on the random forest model,
Alternatively, a request for inquiring whether or not it is necessary to perform a reconfiguration training on the random forest model is transmitted to a predetermined terminal, and a confirmation command fed back from the terminal in response to the inquiry request is received. The electronic device according to claim 8, wherein it is necessary to perform reconstruction training on the random forest model. The reconstruction training includes deterministic training on variables of the random forest model and deterministic training on coefficients of the variables;
The electronic device according to claim 8, wherein the corrective training includes deterministic training on coefficients of variables of the random forest model. As said step S1,
Obtaining a first quantity of service data of a user newly added in a period from the end of previous model training in the service system to the current time, and the first quantity exceeds a first predetermined threshold, It is assumed that the conditions for performing model training are satisfied,
Alternatively, whether or not the model training command has been received is detected in real time or on a regular basis, and when the model training command is received, the condition for performing the model training is satisfied. Electronic devices. As said step S2,
A second quantity of service data of a user newly added in a period from the end of the last reconstruction training in the service system to the present time is obtained, and the second quantity exceeds a second predetermined threshold If it is necessary to perform reconstruction training on the random forest model,
Alternatively, when a request for inquiring whether or not it is necessary to perform reconfiguration training on a random forest model is required to a predetermined terminal is transmitted, and a confirmation command fed back according to the inquiry request is received from the terminal, The electronic device according to claim 11, wherein it is necessary to perform reconstruction training on the random forest model. In the step S4,
S41 identifying a range of coefficient values of variables corresponding to each of the variables based on a mapping relationship between a predetermined random forest model variable and a range of coefficient values of the variables;
Acquiring the value of the coefficient of the variable from the range of the value of the coefficient of the corresponding variable for each of the variables, and performing correction training on the random forest model based on the coefficient of the variable whose value has been acquired;
The electronic device according to claim 11, comprising: A computer readable storage medium,
At least one computer readable command is stored on the storage medium;
A processing device executing the at least one computer readable command;
Analyzing whether a condition for performing model training is satisfied by the model training control system;
Step S2 of determining whether the random forest model needs to be subjected to the reconstruction training if the condition for performing the model training is satisfied;
Performing reconstruction training on the random forest model using sample data when reconstruction training needs to be performed on the random forest model;
Computer readable, it is possible to implement step S4 of performing corrective training on the random forest model using sample data, when it is not necessary to perform reconstruction training on the random forest model Storage media. As said step S1,
Obtaining a first quantity of service data of a user newly added in a period from the end of previous model training in the service system to the current time, and the first quantity exceeds a first predetermined threshold, It is assumed that the conditions for performing model training are satisfied,
Alternatively, whether or not the model training command has been received is detected in real time or on a regular basis, and when the model training command is received, the condition for performing the model training is satisfied. Storage medium. As said step S2,
A second quantity of service data of a user newly added in a period from the end of the last reconstruction training in the service system to the present time is obtained, and the second quantity exceeds a second predetermined threshold If it is necessary to perform reconstruction training on the random forest model,
Alternatively, a request for inquiring whether or not it is necessary to perform a reconfiguration training on the random forest model is transmitted to a predetermined terminal, and a confirmation command fed back from the terminal in response to the inquiry request is received. The storage medium according to claim 15, wherein reconstruction training needs to be performed on the random forest model. The reconstruction training includes deterministic training on variables of the random forest model and deterministic training on coefficients of the variables;
The storage medium according to claim 15, wherein the corrective training includes deterministic training on coefficients of variables of the random forest model. As said step S1,
Obtaining a first quantity of service data of a user newly added in a period from the end of previous model training in the service system to the current time, and the first quantity exceeds a first predetermined threshold, It is assumed that the conditions for performing model training are satisfied,
Alternatively, whether or not the model training command has been received is detected in real time or on a regular basis, and when the model training command is received, the condition for performing the model training is satisfied. Storage medium. In the step S4,
Identifying a range of coefficient values of variables corresponding to each of the variables based on a mapping relationship between a predetermined random forest model variable and the range of coefficient values of the variables;
Obtaining the value of the coefficient of the variable from the range of the value of the coefficient of the corresponding variable for each of the variables, and performing corrective training on the random forest model based on the coefficient of the variable whose value has been acquired;
The storage medium according to claim 18, comprising: